Predicting Recurrence After Radical Surgery for High-Risk Renal Cell Carcinoma: Development and Internal Validation of the "TOWARDS" Score.

BACKGROUND

Considering the reported greater benefits of immunotherapy and its unignorable adverse events in adjuvant therapy for high-risk renal cell carcinoma (hrRCC), accurate prediction may optimize drug use.

METHODS

The primary objective of this study was to generate a score-based prognostic model of recurrence-free survival in hrRCC. The study retrospectively evaluated 456 patients at two institutions who underwent radical surgery for nonmetastatic pT3-4 and/or N1-2 or pT2 and G4 disease. Clinical variables deemed universally available were selected through backward stepwise analysis and fitted by a multivariable Cox proportional hazards regression model. A point-based score was derived from regression coefficients. Discrimination, calibration, and decision curve analyses were conducted to evaluate predictive performance. Internal validation with bootstrapping was performed to correct for optimism.

RESULTS

The mean follow-up period was 55.3 months, and the median follow-up period was 28.0 months. During the follow-up period, the recurrence rate was 48.2% (n = 220) during a median of 75.7 months. Stepwise variable selection retained age, Eastern Cooperative Oncology Group (ECOG) performance status, presence or absence of symptoms, size of the primary tumor, pathologic T stage, pathologic N stage, tumor grade, and histology. Subsequently, the TOWARDS score (range 0-53) was developed from these variables. Internal validation showed an optimism-corrected C-index of 0.723 and a calibration slope of 0.834. The decision curve analysis showed the superiority of this score over the University of California, Los Angeles (UCLA) Integrated Staging System and GRade, Age, Nodes, and Tumor score.

CONCLUSIONS

The authors' novel TOWARDS scoring model had good accuracy for predicting disease recurrence in patients with hrRCC, and the clinical practicability was superior to that of the existing models.